Episode4 : Changing Gears

Key Takeaways

Hi, good morning everyone. Um, so this episode I'm going to talk about uh changing gears. So last three episodes we built a MEP for agentic network application. So now I want to go deeper uh talk about the ways of working and also demonstrate um the different speed at which you can develop with AI. Uh so uh yeah so now what does it mean? uh say when you drive a car let me give analogy uh okay so when you drive a car uh in gear 1 you can go maybe 10 km speed uh if you go more than that uh you will you cannot go right uh so you can ask a simple uh request to AI say for example fixing a logo or login I'll demonstrate that uh so it may generate few lines of code um per it. Second, when you change the gear two in the car, you can drive maybe 20 25 km. So you can give a small feature request. For example, building connections feature for our agentic network. We built profile in previous uh episodes. Now I can ask it to develop a connection to connect two profiles together. Uh so any changes in the back end and uh front end. So that I would say gear two. Uh and when you change gear three, you can go maybe 35 maybe up to 40 um kilometer per hour, right? So now I'm going to take an API for that matter. Uh so I want to give a specification uh which I created. I'll show that how did I create the specification. I want to introduce the agentic core for uh our agentic uh networking professional networking. So these are the three topics I'm going to talk about in this episode. So that's why I named this named it as a changing gears. Uh so let's see what it is. Yeah. So let me move out of EP and uh let me talk about those features development. Uh today I'm going to do a live development. Okay. So now let me just check everything is fine. Yeah. Yeah. So now what I what you see in the screen you see only console right? Uh so in that console uh what you see just the command line prompt I created two command line prompt to start with. So let me show how do I start development. So I start my plot CLA. Okay. So maybe I'll just uh warm it up. Okay. So, let it uh maybe go through all the document again. Around 20,000 lines of code is there. So, it may take a few minutes to understand. In the meantime, I'll just uh log to the agentic platform that we created and uh I'll talk about uh the features that we are going to develop. Uh okay. So now I logged into Agentic network which is a live version uh which is running in Firebase and Google Cloud. So when I log in uh I see okay this is the dashboard that I landed on. So I'm just checking as a maybe as a tester. Uh you see the logo is not uh set. But if you look at uh Firebase or uh cloud it is having some logo. So that is a one small uh user story I'm going to take it and it may take few minutes to fix. And second thing last episode we saw there are some glitches in the login. So those two I'm going to do the fix in the gear one. Right. Uh so these are the two things I'm going to fix it. Uh so let it uh get to worse with the code. Second feature uh the gear 2 feature say right now we created a profile. So you can go and view the profile. Uh so when you join the network it doesn't show. So right now also when you view the profile it is showing is a logged out version kind of thing. So I'm going to build second capability. First capability that we built is the building login and user profile creation. Now third feature which we are going to put it in the application is the connections. Okay, that is a feature not a part. Okay, now third part is very big. So I'm going to introduce agentic core. For that I did some analysis. So I'll show how did I if I want to build epic it simple prompt will not work. It may work uh but it will create something that you might not uh intended to create. So for that you need to do some homework. Uh so it is not the typical way of u feasibility analysis or having a product owner and uh uh doing IP sprint uh in your safe execution etc. It is all going to be in minutes only in the future. Uh right I'll tell you how did I explore uh so last week after posting the episode 3 I was thinking maybe it's time to introduce a a feature and um agentic part so I went to cloud in the UI uh say cloud cloud UI web interface so let me just explain uh okay first I asked okay I I'm building a uh professional networking uh application. So I want to put a first approach um and then it started uh thinking more even I did not give proper com prompt. Uh so then it realized okay this is a fascinating design question. So then it uh started talking about okay um connection uh rethinking a connection what connection means agent to agent handshake qualified lead uh for connection uh or it could be for business uh human endorsement connection. So then it started thinking and it started putting architectural shift uh delegation and permission model agent reputation trust score and human checkpoint etc. though then it started um uh thinking right then I asked okay what are all the agents I need to build because I have some thought but I want to structure my thought uh with AI. So I asked what are the different agents and what are the different purpose if I want to build agent first networking platform and how all of that should be orchestrated because now agentic uh there are a lot of uh different thoughts u you can build agentic workflow you can do multi- aent or you can have single agent focused and task based agent so I wanted to pick the brain of AI which is uh shown at 4.5 uh the latest models from anthropic so then it started thinking Uh okay it took some time to do a analysis etc. Then it said agent social graph uh before it all networking uh professional networking graph. Now it after that it said professional agent personal agent profile agent outreach agent gatekeeper agent for each agent what is the purpose and responsibility. So it gave a set of agent platform side agents and then it gave the architecture. Okay. how uh you need to set the goals and preference. So in professional networking the goals would be different for one person the goal could be to get a job. One person goal could be to build the network to improve the skill set. Uh for a engineer it's a evolving skill right you may want to connect with the agent. Yeah this completed we'll come back to this quickly. So it uh give a proper um uh control plane orchestration plane uh right and uh agentic core with the agent and agent to agent protocol and platform layer etc. So now like that it gave lot of uh uh inputs I was having some conversation and gradually I built a very good uh specification. So there is a specification that you see on the right hand side which I downloaded uh which I'll be giving I'll open that in my uh local laptop um and show you. So these are the three things that we are going to do. So now our clock is warmed up. So we'll go get into actions. So if you see this is the complete specification. Uh, okay. But I'll download it again. Okay. So now uh this is a specification I'm going to use for the third uh gear three. So that will build autonomously everything. So if you see the specification look at the uh clarity of the specification even human cannot write. Uh we'll maybe spend couple of minutes here. So it has given agent core technical specification. So I tweaked it bit. So it created a firebase version. uh it given a executive summary uh right and then it has given technical stack uh given input what is the MEP that I built on I'm using react uh and then Firebase uh for the back end and uh MEP I already completed for profile creation so next release content it is creating that agent decore function etc. So industri firebase architecture and then uh technology stack agent framework llm integration I am going with the GPT5 mini because that is a cost effective model and better model at that cost. So I don't want to use cloud 5 for now because it's going to cost me more. I will use cloud sonet 5 for coding. Uh but for lm integration maybe I'll stick with low cost model. later on uh I can uh use some different model for reasoning and some model for low task uh simple task. I'll change the model based on the task that I do. So it has created very good uh specification. This is what I'm going to give it for the epic uh development. Okay, let's uh start building it. Uh so now where is my plot? Yeah. Okay. What uh so it understood the code base. So, so we don't need to uh read through. Uh, so now it is asking what you would like me to focus on today. So, now let's fix the uh logo first. Uh, okay. So, I do not see a logo. I do not see a logo for the application. Please help me. Uh so help me to create a nice full logo aentic networking platform and save that in the application. Slow down. Yeah. So this is user story. This is the first one and second is the login issue. After that we'll go to the feature de. This is the gear one. Okay. It's it's minutes or seconds. Okay. So, it's first focusing on cing uh before creating the logo, it's finding out is there any already a logo. It's very cool. Uh so nowadays I think you need to know what to ask not to know what is the answer. That is the uh new era um that is coming. You don't need to know how to uh go and uh where to create the where to put the logo so that it appears in the browser. So you just need to ask what you want. Yeah. Now we started working on uh designing and creating SVG SVG logo. Before in the traditional world we need one designer or you ask for a logo creator. So you you will do some iteration and you will come back with some logo. So it created for now I'll keep a control and give up roll for every uh change and later on maybe when we go for epic I'll give full control. Yeah, now it's creating the favcon files which will be um loaded in the browser icon. So you don't need to be a front- end developer, you don't need to be backend developer. All you need to know how do you work as a a engineer. Uh you need to know basic uh basically you need to know the breadth not the depth anymore. Uh so new skill Yeah, in the future you need to know only breath depth AI will know in the tips t-shaped skill or p-shaped skill. Uh the depth will be a responsibility breath will be the only the human responsibility. Yeah, now it's copying the uh Favicon file into the right location. So it's creating a component uh properly u so that it can be reused in many pages. So you can relax nowadays. Uh the the load cognitive load on developer will not be high in the future. So all you need to know say the coding will be done by a you can relax and then you don't need to focus every line and then do the coding for every character. So you just need to kind of put it in auto cruise and then sit Relax. it is making any changes in anywhere else. So typically when you make a change you need to do the impact analysis and then make the relevant changes. Uh cloud code is very good at it. So it never misses any uh related areas. So, it's adding the logo in uh sign in page. there. Now it's heading into signup page. I don't know whether it added in dashboard page. Let us see. So look at the way it developed. First it developed uh it created a um logo and it created a component uh logo component and then it is adding that component in all the pages. Yeah, now it's saying let me test it. Okay. Yeah, this is this is what it's interesting when you do a developer. So after all of that he will build. So now it's building now. So automatically it will code everywhere and it did the impact analysis and did the changes in all the places. Now it's trying to build to see is it's getting any typescript error uh any not imported import error etc. So any syntax error so it if it finds any error it will fix it automatically. You don't need to ask for the prompt. So it it uh clot code has engineered that u uh way of uh doing autonomously uh coding and then building uh testing testing it's not unique testing the basically the building compilation error and making sure compile free code is uh generated and most of the time 99% it doesn't create any um error. Okay. Okay. Now it is saying it's ready. Uh so it will give a summary uh what are the changes it will do it has done. Uh so it created um a logo and it updated in uh login landing page, dashboard page, sign in page, sign up page and link complete page. Okay. So now So if you don't know the command how to deploy you can ask say I use 23 firebase project uh to test uh so so I want to make sure I use the right project. So it is saying uh okay now it is uh so now it it understood that since I asking a command so I need help in deploying so it may deploy by itself also so sometimes I keep the deployment under my control uh otherwise it will simply deploy for every change but let me just give the control So to make sure the fire right firebase project is used, it is uh uh checking u firebase project list and then uh it is u setting the agent uh firebase it is using agentic network. It's very smart. It's smarter than a senior developer. Now it's checking everything before deployment. Perfect. Your configuration is correct. Uh here is the verification and deployment guide. Okay. Yeah. So it is using aentic network. Now deployment guide. We need to do Firebase deploy hosting. So let me do the Firebase build. when you don't select specifically it will deploy everything. So let me build everything and deploy only printed If you have any questions, you can put post it in the comments in YouTube or uh LinkedIn. So I'll try to answer deploy is also building. So maybe next time we don't need to build before. Yeah. So now what we'll do, we'll skip the after this we'll skip the login issue that I'll do later. We'll go to the connection feature and uh the autonomous uh agent core. Okay. So now we deployed. Let's see. I think I'm curious to see what is the logo it created. So right now here here you see there is no logo. Okay. So now let me refresh. Yeah. So see see put it thought through what should be the uh icon. So since it is networking so uh the person uh can make a effective connection. So it generate a very nice icon. Even a logo creator wouldn't have thought through this uh purpose. Okay, now this is done. So how long it took? Maybe 20 minutes, right? So now I did not do any coding. So now this is done. So now let's go to the second feature. So now which is the second feature building a net uh connection, right? So now I'll give it an autonomous uh capability. So now let me yeah let me appreciate now we need to build connection feature. So I returned some prompt I just copy pasted. Uh so basically I'm saying now we have created the profile feature. Uh or we we I think I'll simplify I'm saying too many unwanted words. Uh so I'll simplify the prompt. Basically I'm saying I have the profile feature. So I want to build connections feature. So uh okay let it be. So now that we have created profile feature I want to build connections feature. Uh please analyze the codebase and make necessary changes in the back end and front end code. Make sure you think architect design well before start coding. Uh and choose the good design principle. For example, keeping all the user data under user collection discoverable profiles. Keep it separate so that uh data uh security is uh appropriate. Okay. So let me give this command and um it will do uh in the meantime I'll give complete approval then I'll explain certain other things. Yeah. So now this is one new feature that is see clock code entered a mode called plan mode. So whenever it need to think it goes to the plan mode. So, so now we have given clear instruction that you need to think uh architect design before you start coding so that you do it wisely. So, it entered into plan mode. So, in a plan mode it will think more and then it will design and then go to the coding. Okay. So, right now it is in plan mode. It enter the plan mode to design the connections features. Let me explore the codebase to understand existing architecture design based approach. Okay. So now why I ask that uh detail uh right right now I have only profiles collections. So as we build more feature more collection uh more uh documents will get created. Right now I have profiles which is uh public information which are discoverable. Uh right I want to keep public discoverable profile separately and user content separately. how many connections he's having until he gives permission for that discoverability that should not be exposed right that is why it is better to design uh in profile we have only profiles as a collection as we build more feature you will have more uh collection and it will grow up for enterprise level um global networking platform right now it is having very simple profile structure where it has um the profile detail education experience etc So right now when we build collections then we'll see what are the collection structure after we add a uh uh one profile connected to another profile. In the meantime I created two user right now in the database you saw two profiles right uh both are mine uh one with one email id and another with another email id. Mohamad rafi is one and mohamadi yes is one right. So both I have logged in. Uh so now let me log in. Yeah, this is the glitch I need to fix. I think I'll fix this later. I'll create one. H. Okay, it logged in. Okay. So now these are the two profiles I have. One uh you can see with the profile photo which is uh Mohamad Rafi. Another one is Mohammad Rafi. Yes. Okay. So now these two I want to use the connection feature and make both the profile connected. So that is what feature we are building. So let's see how it is building. Uh in the meantime until it comes back uh it is doing some work in the back end. Uh we'll look at uh the agentic core um specification because that is very important although it created autonomously. what I'll do I'll make this uh yeah so we'll keep checking whether it is completed okay so now let's look at the document which is very good to understand because now you need to build application in a reimagined way what it means is you will have uh we'll go to the architecture and then talk about which is very important. Every person a engineer need to be aware because I see there are a lot of uh different uh thinking and perspectives happening. Somebody is thinking I need to use AI for only it which means how can I build faster the way I build it. Okay, other crowd is thinking okay I need to build agentic application leaving aside how business is working currently uh what is needed or what will happen uh right it is a mixed approach you need to have uh application uh which is uh kind of working in the current ways of u uh building the application what it means deterministic application okay it is not here maybe I have it in the design specification and you need to build agentic features. So both need to be interconnected. You will not have separate agentic application sitting outside and you will have traditional deterministic application sitting uh separately. Uh you need to reimagine how the IT application need to be built with agentic first approach. Um okay let me open that um architecture give me a minute yeah so here if you see connection data model already defined wind types so it is working on credit etc let it work okay in the meantime we'll I'll teach you something else so because Uh we don't need to uh yeah okay it's asking some question create separate user collection yes so whenever it is having question in mind so it's like a developer comes back and ask an architect or lead developer uh so now it is asking uh why because the profiles is the only collection uh it is there right I instructed you need to keep separate so it is asking option one you want user collection separately or you want to put it in profile with security it is also giving the recommendation what it thinks so I'll that is the right option uh then second question search browse users by name skills location send receive connection okay now I'll select all so see see the different types of questions so now they're making cloud code more effective to be user interactive whenever it needs input it is asking. So first is a single uh option. It is a multiple multiple selection now. So I want user to filter uh when you connect based on name, skill, location and uh connection request and uh we manage connection list agent initiated connection request. Uh so yeah so I select okay wait and next. How do I do? So these are the okay let me complete and talk uh birectional mutual like LinkedIn privacy hide email from public preview I profile from search unless connected H. So now it took input and then it will start working. So you given a a prompt and then it started thinking and working in between the working mode it is if it needs an input uh to make it more effective it is asking so before we started with w coding and then a spec coding with more detail now I don't know what do you call this one it's a interactive coding uh so within the w coding so it ask a question uh and it collaborates it's a collaborative w coding okay so let's come back to our agent reimagined uh thing. Right. So now this is the base layers. Okay. So now let me explain with this view. Okay. You will have client layer. So which means you need to have UI uh to visualize things which is uh which will be combination of two. you will have conversational plus u navigation. So user interface will before it was only navigation. So in the future it will have conversation plus navigation and plus autonomous. So which means agent will send request to another agent and it will accept if it is I autonomal level setup. So which means no conversation at all. Uh if you want to have conversation you can have a conversation. If you want to have navigation, you will have navigation. That UI layer, client layer will be having uh navigation plus conversation plus autonomous. So that is the change that will happen. And in the business layer uh before we used to have microservices and API uh which are used to uh which are uh called by the client uh so back end for front end uh we add so that the client are more efficient. So now in this layer you will have agentic core and orchestration uh layer and a lot of agents uh and you will connect within the agents in your application and you will also connect agents across the enterprises. So that is the world coming. Uh so now you will have single agent, you will have multi- agent uh you will have agents within your enterprise, you will have agents across the enterprise. So agentic mesh all the concepts are coming in. So now I just want to focus on that part. For each agent, you will have agent skills and you will have uh LLM API to be called passing on the right context and right prompt to get the right uh response and stitching that stateless LLM API call into memory based and taking the decision based on business logic etc. Now let me explain to the agentic core. Okay, agentic core. uh right now. So it is this is the initial uh when we created the MEP uh so that time told about agent core components. So basically here you will have multiple agents uh one for person here it is depicting personal uh we'll go to the this pick after this. So here in the initial view it has given two three as an example. Okay. Personality engine. So tone, style, strength uh and then uh which can be used uh and the context coming from interactions and user preference and goals. If the goal could be uh getting a job if a person coming out of college or if a person existing working in IT for some time you need to reimagine uh to a engineer. How do we build with AI? How can you build agentic application? So you need to get a similar skilled people in the network so that you can improve skill faster. So the goal could be different. Uh and uh decision engine uh how can he approve a connection request etc. and memory system you need. So all of this the agent core is what you need to reimagine your application put into your traditional deterministic system. That is the key part to observe. Okay. The layers will change the client layer UI layer will change to have navigation plus conversation plus autonomous and business logic layer will change before microser API is there. Now you need to have agentic core plus microser API plus tools for MCP server plus agent skills. Okay. So that is what we are going to build as API uh after this uh connection things built up. Now we'll go to the specification which was was created uh to build this capability which are more detailed I think it's asking. Okay. This allow what it has done. So now it has detail plan to create must create files, must modify files and security checklist etc. Then what is asking migration and security? Okay, this you forget about it. I think it's doing some traditional estimation technique. It was thought to do. You don't need weeks etc. Approve this plan. Confirm architecture approach. Uh okay. Now the architecture priority security. Okay. Yes. Not accepted it. Yeah. Yeah. Now it has completed the plan. So it is going to uh build now. So it has a clear view what are the files it need to build. Um and then it will start building all of this uh one by one. And it also thought through it need to make a database changes for collection and indexes rules for access etc. So process profiles collection uh okay let it complete let it complete and the meantime I'll explain the agent core which is interesting part which you need to rescale Yeah. Here it has put gatekeeper agent, uh, profile agent, orchestration agent, etc. Uh, so let me keep it like this. Yeah. So, trust agent, scheduled, etc. Right. This all h so from this we'll see uh agentic implementation. So now there are different frameworks. Okay, there are a lot of agent kit, development kit. There is one from Anthropic, there is one from Google, uh there is one from AWS, there is one from OpenAI. So now I use longchain. Uh so that is open source. So maybe I'll go with that. Yeah, the agents profile agent it is guiding us to do based on uh fire store trigger gatekeeper agent and orchestration agent and it is using different techniques uh when that agent will start. This is agent to agent conversation. So you have to test these are all new parts right? Some places it may work perfectly some places we may need to uh change uh after the base implementation. Some places it may fail also because these are all not proven thing. It's evolving every day uh and every week uh right. So now we need to start building the skill as it uh it also given cost estimation since I'm using Firebase. Uh so I need to move to Bliss which I already did. So it may cost me 400 users uh to run this application anywhere between $5 to $20 per month uh for cloud function and uh cloud fire store. It may cost $2 to $10 per month uh 400k read and write etc. and firebased hosting uh I think and open a cost uh it is given the estimation for the MVP phase also. So GPT mini is used for certain task and GPT 5 is used for certain task for thinking. Uh so the cost is different. So if you see here for 50 conversations per day this is going to cost uh 75 100 uh conversation per day this is going to cost 30 uh because the mini will cost uh lesser GBT 5 is a bigger model so it will cost um more but we can use it for reasoning task etc. So overall cost $110 per month for 100 user. That's what it is saying. So that's okay. I'll spend that and then uh maybe I'll it will help me to learn to build agentic application with agent core. That is the purpose uh that I'm doing this episodes and you can also learn but you will not get a realtime experience until you build an uh application with agentic core into that. when you do then you will start thinking until then your mind will not think uh so you need to start building a aentic core agentic application then only you will realize okay whether I need to have a use case just by calling LLM uh don't talk too much of buzzwords of agentic multi- aent etc even that works very great uh if you give the right context uh and then do the right prompting uh then you can uh get right results. You need to balance between deterministic system. Okay. It is done. Excellent. Let me make the phase one complete. Data migration security successfully implemented. Foundational architecture for connection feature has been implemented. What I'll do? I'll delete uh both the profile. Okay. I'll close close and then and I'll delete the profiles. uh collection will start fresh because now it's going to create profiles and collection user collection object. So that is a change it is sayingflow will create uh user and profile documents on signup. Um okay we'll delete the old MEP u data which is any testing phase only right now today's session may be a little longer uh because I'm planning to cover the agentic part also let us see yeah now when it builds it got the error so it will fix it by itself Uh, I need to fix a compilation error. The dashboard is still trying to use the email field. So how long it took another uh 30 minutes right? Uh approximately. So in 30 minutes you can build a feature. So before you cannot even detail the feature in 30 minutes. And uh please note that uh right now the connection and LinkedIn are the general terminologies professional networking was there for um so many year it is public domain uh right uh but if there is any specific domain where uh the detailing of feature is based on specific uh knowledge which is not known to LLM then it cannot do. So you need to be uh you need to know that part for that you need to have a domain modeling domain based llm maybe you need to fine-tune the model etc but uh on a general context building any application uh building front end and back end the technology part as a developer it will do if you want to build anything specifically the domain knowledge you need to give give it more detailing in the specification or as a requirements in those cases I think it may be taking a little longer to write the specification. So basically context engineering will take longer time but uh your uh coding will not take longer So once you have this skill to develop uh with AI way to how to work with AI to run um with all the techniques and to do gear 1 and gear 2 gear 3. So basically you will become very powerful a engineer uh who can build application at a rapid pace uh which which is the which is the highly wanted skill. Now who are all having that skill they are launching um they're part of the enterprises or they they can create their own company with small team uh to build a unique uh idea quickly into market. The innovation there is no limit now. So the pace is uh very high uh and you the dependency on technical capability is kind of thrown out uh right. So you can go at faster pace. Right? Now you need to reimagine every industry need to reimagine otherwise there is a risk uh for existence of that particular company and particular industry competition is going to go crazy. Yeah, now it built it is deploying once it uh when once the connection feature works I think logically we'll conclude this episode. Maybe I'll start episode five for agentic core and uh maybe both I'll publish quickly. Yeah. Yeah, now it's doing one by one. First it deployed indexes etc. Now it's deploying the back end cloud function and it will deploy the front end. Front end it deployed already. It deploy. No, it just built it here and you need to use serverless um serverless architecture because uh the agentic whenever it needs it will use the computer. So you don't need to it chooses the serverless architecture. If you see uh if you manually code people will do um either um only microservices and uh containerized replication. But now it uses the best whenever we know use case or whenever agent need to work then only the compute will get kicked in. Yeah. So this is the function is getting deployed. So this is the agentic network. Uh so database we will delete that once it completes and hosting is what I use uh for the agent network app. It's also verifying those functions are there running. Please share your feedback. Uh so I just started uh this episode series to upskill um um your capabilities. So if it is going in right direction, let me know. Uh if we need to improve, if we need to focus on anything, if something is not clear, uh let me know. So let me double deploy the front end again just to make sure. Okay. So now what is saying I'll delete this collection. Let me also check the Google sign in. Is there any challenge in the way it is handling the authentication? So based on the prompt it will go to the right file. That is the beauty. Uh it understand how to navigate even if you have thousands or more files. Okay. It found some issue issue identified. Let us see. I don't know. It will identify the context and it will figure out the issue and it will go and change. Okay. Most of the time it works. Um for a human to triage it's very difficult. Uh actually normally first you need to know which file I need to change. If there are too many files until otherwise you have everything in your fingertip uh the structure and the files etc. Uh then you need to debug. Uh you will do uh some testing and then you will fix it. Then you will test it. Sometimes you need to be cautious if your code is more stable. Uh then if you want to do cautiously change, you need to go to that particular file and uh narrow down the context and then work on that file otherwise it will just change everything. So you need to be careful when to use gear 1, gear 2, gear 3 or stop uh if needed and take the control back and maybe you need to do manual coding at times uh if needed. So in this episode we'll be covering the gear one and gear two the agentic core. I think after concluding this episode I will kind of start on episode five uh which is agentic core. Yeah. I'll publish the agent core episode 5 tomorrow. So let's uh conclude on connection features today. So now that I deleted all the authentication and user, so it's going to be new sign in and uh I'm going to set up two user and then connect to to each other. 11 a.m. Okay. So now let me explain this also. Say I use cloud code. Uh uh I will explain the subscription model now. So now say I use too much of token and cloud code is putting a restriction. So it is saying okay you reach the limit either you wait until 11:00 a.m. uh which is uh for me it's 9:00 a.m. almost 2 hours or uh you upgrade to next plan. Okay right now uh let me go to cloud what it means I'll show you. So cloud uses two model one subscription plus uh API. Okay. Now if you you need to have uh you need to have uh pro pro subscription then only you can do cloud code that is number one. So the pro is having certain limit. Uh so after you enter limit it will tell you okay either you need to wait come back your uh um uh usage limit when it enabled again. So right now I hit the limit because I'm using too much of token. So now it is asking me to either you upgrade to max or you wait. Okay. So now for me let me upgrade to max. Uh uh I don't want to wait. It's okay. I think I'm learning my skills. So I'll spend some cost uh for the learning. Let me upgrade it in another screen. It's encrypted only. Yeah. So, it is giving 5x more usage uh than pro. So, right now whatever I use it will give me uh 5x more capability. If you want 20x more cap uh token usage then you need to upgrade to 180 plan. So they have uh $15 plan uh which is pro and uh max uh is this and then this is your u super max or I don't know whatever they call. So I think for me I'll go to $9 since my card detail is there. I think it will go nowadays everybody's becoming sponsor uh I think in a single click it will allow me to upgrade. Done. Okay. So now I can continue. Okay. right now. Yeah. So, I've been spending some money but basically I'm learning my ski my skill is improving. So, that's all matter. I'm using this cost as a training cost for me. So, let me go back uh and then say cloud port is sending to upgrade. So, basically they're doing upselling. in cloud code. Uh so see the see the beauty. So that's why claude and anthropic is becoming more profitable. You can you can look at their uh uh how did they do in 2024 and how how they are doing in 2025. They are kind of grown up uh in their revenue uh AR uh drastically in this year. Yeah. Now let's come back to our our work. Yeah. What it means if you want to start first you need to have cloud pro subscription plus API. I think API you may need or not don't need. Write down exactly uh without CL cloud pro I think whatever $15 the basic subscription cloud plus pro yeah pro pro right so you cannot do cloud code u so you need to have that basic uh $15 subscription then you can do cloud code and that is good enough for you to start with uh so I used that for 2 3 months and after that I went to pro then I felt it is too much 5x I'm not using. I went back to pro and now I'm again hitting the limit and then changing it to max. So you can find out what is your need and accordingly you can keep either pro or max for individual user. So enterprises I think there could be different plans that I have not figured it out. Uh but for individual uh to gain your skill, I think you may need to invest $15 a month to make it hands on and then use the tokens wisely. Don't waste it. Think of a good idea. Don't follow somebody and whatever is interesting you pick up an application and start building and feel how a is working and uh do with the gear 1 gear 2 gear three approach something with the slow approach uh asking at the context in a file something at the modular level feature building something at the API level then you will gain your a engineer a empowered engineer capability Now it's I think started moving to phase two. What is phase two? I think it clear since we put max. I think it's starting to accelerate to phase two. I wanted it to stop and if you give right specification it can code up to I think many hours. I have seen it coding for 30 20 minutes continuously. Uh it can also code um in hours together. Uh that's what I'm hearing in some of the videos from anthropic. Uh so it has a capability to go code continuously. Let's see if logically if it stops somewhere we close this episode. So it's creating the UI component functions and components under connection is creating many components. is going to code here now. So let's create a page and finally we'll see how many what is the lines of code in our project after this. So it created the components and it created the pages and then it's creating the request uh connection request and uh acceptance approval details. So these are the modules. So it's structured properly. Profile is a one module. Dashboard is one module. Connection is one module. Authorization is one module. Authorization you have sign in, sign up. uh connection it is having request sent so it properly structures which is I think amazing people ask okay is it hallucinating or not I'm amazed to see uh what it is doing currently Just trying to build some hoops. see it's already yeah it took around 8 minutes uh continuously to code. Okay. So now what is the summary is done phase one phase two is done. What's working now? What's next? search discovery. Yeah. So what I'll do logically I'll close this episode here but I'll let it run to complete phase three phase 4 phase five in the background and in next episode what you will see is building the agent core functions. Uh so until then uh this is Rafi. So see you in next episode. If you have any feedback, please uh let me know and uh either in YouTube's YouTube uh channel video comments or in uh LinkedIn uh comments uh or you can send one to one uh message to me in LinkedIn. Yeah. Thank you. Happy learning. Bye.